Data-driven planning and truncation methods, devices, and storage media

Abstract

This invention discloses a data-driven planning and cutting method, apparatus, and storage medium, relating to the field of coal mining planning technology. The method includes the following steps: acquiring a set of phosphate rock layer image data; preprocessing the images; and manually labeling the rock strata, phosphate rock layers, and cutting boundaries to form a training dataset. A detection network model is established based on this dataset and trained using labeled images. After real-time acquisition of mine images, these images are input into the trained model, outputting labeled images of the ore layers to be segmented, and extracting and optimizing images of the fully mechanized mining face. An adaptive grid size is determined based on the breaker radius and roadway cross-sectional area data. The cutting path is simulated based on raster map data, generating and merging path segments to obtain a comprehensive cutting path, thus completing the planning and cutting method. This improves the continuity and efficiency of the mining process and reduces overlapping mining and omissions.

Description

Technical Field

[0001] This invention relates to the field of coal mining planning technology, specifically to a data-driven planning and cutting method, apparatus, and storage medium. Background Technology

[0002] In modern mining, especially in phosphate mining, accurately and effectively identifying and planning ore seam cutting has been a long-standing challenge. Traditional methods for ore seam detection and mining planning rely heavily on human experience and simple image processing techniques, which suffer from insufficient accuracy, low efficiency, and safety hazards. The complexity of ore seams often renders manual annotation and experience-based judgment inadequate for practical needs, resulting in insufficient scientific rigor and reliability in mining plans.

[0003] In ore cutting planning, grid division often lacks a scientific basis. Due to the complexity of geological conditions, the fixed grid size in traditional methods may not be suitable for the actual situation in different areas, resulting in some ore not being effectively mined. For example, if the grid is too large, it may fail to capture key geological features; if the grid is too small, it may cause information redundancy and increase the complexity of processing and calculation.

[0004] In the prior art, CN112883559A discloses a planning and cutting method and device, storage medium, and electronic device based on a big data system. The cutting method includes the following steps: acquiring data on various equipment and geological information of the longwall mining face collected by sensor devices and establishing a geological model; calculating the data conversion from the geological model to the planning and cutting model through a deep learning neural network of big data, while integrating the attitude data of underground equipment returned by the central control station to calculate the actual underground position of the coal mining machine, the exposed coal seam thickness, and the actual mining height; using the calculated data conversion to establish a real-time planning and cutting model of the coal mining machine, support, and three machines; and using the established planning and cutting model, combined with big data decision analysis of multiple sensors on the working face, forming a planning and cutting curve, and sending the planning and cutting curve to the coal mining machine control system, which then performs automatic cutting according to the planning and cutting curve. However, the planning and cutting curve formed by combining multiple sensors on the working face with big data decision analysis does not involve specific analysis of the working face. Unreasonable grid division and mining path planning lead to overlapping mining or omission of ore, directly causing waste of resources. This affects economic efficiency and the recovery rate of mineral resources, thus reducing the accuracy and effectiveness of the cutting results.

[0005] The information disclosed in the background section is only intended to enhance the understanding of the background of this disclosure, and therefore may include information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0006] The purpose of this invention is to provide a data-driven planning and truncation method, apparatus, and storage medium to solve the problems mentioned in the background art.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] A data-driven planning and truncation method, comprising the following steps:

[0009] Acquire phosphate rock layer image set data, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images. Store the labeled images and the corresponding original images in the database to form a training dataset.

[0010] Based on the data in the training dataset, a YOLOv8 detection network model is established. The original images of phosphate rock mining without labels in the training dataset are used as input to the YOLOv8 detection network model, and the corresponding labeled images in the training dataset are used as labels to train the model.

[0011] Real-time acquisition of target mine images is input into the trained YOLOv8 detection network model, outputting a labeled mining image of the ore layer to be segmented. Based on the labeled contour of the output image, the fully mechanized mining face image is extracted, and its region is divided by an initial grid to obtain a fully mechanized mining face grid image.

[0012] Based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, the initial grid size is optimized to determine the adaptive grid size. The fully mechanized mining face image is then divided into regions according to the obtained adaptive grid size to obtain raster map data.

[0013] Based on the obtained raster map data, the cutting path is simulated, each raster is traversed, several path segments are generated, and the generated path segments are merged to obtain several comprehensive cutting paths, thus completing the planned cutting.

[0014] Furthermore, the image set data of the phosphate rock layer is acquired, and the acquired images are preprocessed. The preprocessing includes image dehazing, image enhancement, and denoising. Wavelet transform is used to denoise each phosphate rock layer image, and bilateral filtering is used to enhance each phosphate rock layer image sample.

[0015] Furthermore, the rock strata, phosphate rock strata, and the truncated boundaries between them were labeled on the preprocessed images using a manual annotation scheme. The specific annotation logic was as follows: the open-source software Labelme was used to label the samples, the dataset was converted into JSON format, and it contained two categories: phosphate rock strata and rock strata. Each image had a corresponding JSON annotation file, providing object outlines and corresponding classification labels. The dataset contained a total of 2584 images, which were divided into training and validation sets in an 8:2 ratio.

[0016] The method for generating the training dataset is as follows: the labeled images are stored in the database and mapped one by one with the images labeled with the corresponding labels to form a corresponding grid, and the resulting grid is recorded as the training sample dataset.

[0017] Furthermore, based on the data in the training dataset, a YOLOv8 detection network model is established. The training method for the model includes a freeze phase and a thaw phase. Finally, the difference between the loss of the training set and the validation set is within 0.016, and the training ends. The input of the trained YOLOv8 detection network model is the target mine image, and the output is an image with the rock strata, phosphate rock layer and the cut boundary between them marked in the image.

[0018] Furthermore, based on the acquired hammer crusher radius and area data, the initial grid size is optimized to determine an adaptive grid size. The formula used to calculate the adaptive grid size is as follows:

[0019]

[0020] In the formula, To adapt to the grid size, This is the initial grid size. The factor representing the influence of the damage radius is... This is an index representing the impact of geological complexity, among which... This is the weighting coefficient for the influence of the damage radius. The weighting coefficients for the geological complexity influence index are as follows: , and and All are greater than 0;

[0021] Among them, the influence coefficient of the damage radius The formula used for the calculation is:

[0022]

[0023] In the formula, For reference, the radius of the hammer crusher This is the actual hammer crusher radius;

[0024] Among them, the geological complexity influence index The geological complexity influence index is calculated based on the cross-sectional area of ​​the roadway in the fully mechanized mining face and the hardness and number of cracks in the hard rock strata within the roadway cross-section. The formula used for the calculation is:

[0025]

[0026] In the formula, This refers to the cross-sectional area of ​​the roadway in the fully mechanized mining face. For reference, the cross-sectional area of ​​the tunnel, This represents the average hardness of the hard rock strata within the cross-section of the fully mechanized mining face roadway. This represents the number of cracks within the cross-section of the tunnel.

[0027] Furthermore, the fully mechanized mining face image is divided into regions based on the obtained adaptive grid size to obtain raster map data. The specific steps for obtaining raster map data are as follows: the fully mechanized mining face image is divided into regions using the determined adaptive grid size to form several grids. The annotation boxes of phosphate rock, hard rock layer and the cutting boundary between them in each grid are analyzed. Phosphate rock is marked in white and hard rock layer is marked in black to indicate the presence of obstacles. Grids with a black area ratio of more than 40% are marked as black, and grids with a black area ratio of less than 40% are marked as white, thereby generating a raster map.

[0028] Furthermore, after obtaining the raster map, a Cartesian coordinate system is established, with its origin O located at the upper left corner of the map, the x-axis pointing downwards, and the y-axis pointing to the right. A matrix of the same dimension as the raster map is created, and the raster to be traversed is represented by the number 0, the traversed raster by the number 1, the raster marked as black by the number 2, and the raster not marked as black by the number 3. Let all the raster cells on the map form a set. Then the grid cells to be traversed can be represented as:

[0029]

[0030] in, This is the row index of the i-th cell in the established Cartesian coordinate system. This is the column index of the i-th grid center point in the established Cartesian coordinate system. This indicates the traversal state of the i-th grid cell. This indicates that the grid cell is to be traversed, where For the index of the raster, , This represents the total number of grid cells.

[0031] Based on the obtained raster map data, the truncation path is simulated. Each raster is traversed to generate several path segments, defined as follows: a path segment is an ordered set of all connected raster cells in the same row on the map that are not blocked by obstacles. A path segment in column R then satisfies... Specifically, it satisfies:

[0032]

[0033] In the formula, This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are no obstacles in the grid at the row position. The difference between consecutive raster indices within a path segment is 1, used to ensure they are consecutive. For the index of the vertical raster in column R, This represents the total number of vertical grid cells in column R. For the index in column R, excluding the first vertical raster;

[0034] If a certain region of a map is divided into n path segments, then the set of these segments is... And satisfy:

[0035]

[0036] In the formula, Let f(x) represent any two sets of path segments, provided that they have no intersection and are independent of each other.

[0037] The present invention also provides a data-driven planning and truncation device, which is used to execute the above-described data-driven planning and truncation method, comprising:

[0038] The training sample processing module is used to acquire phosphate rock layer image set data, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images using a manual annotation scheme. The labeled images and the corresponding original images are stored in the database to form the training dataset.

[0039] The prediction model segmentation module is used to build a YOLOv8 detection network model based on the data in the training dataset. The original images of the unlabeled phosphate rock mining layer in the training dataset are used as input to the YOLOv8 detection network model, and the corresponding labeled images in the training dataset are used as labels to train the model.

[0040] The planar image processing module is used to acquire images of the target mine in real time and input them into the trained YOLOv8 detection network model. It outputs the mining images of the ore layer to be segmented with labels. Based on the labeled contours of the output images, it extracts the fully mechanized mining face image and optimizes it to obtain the fully mechanized mining face grid image.

[0041] The grid size optimization module is used to optimize the initial grid size based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, determine the adaptive grid size, and divide the fully mechanized mining face image into regions according to the obtained adaptive grid size to obtain raster map data.

[0042] The truncation path synthesis module is used to simulate the truncation path based on the obtained raster map data. It traverses each raster, generates several path segments, and merges the generated path segments to obtain several comprehensive truncation paths, thus completing the planning and truncation method.

[0043] The present invention also provides a non-volatile computer-readable storage medium containing computer-executable instructions that, when executed by one or more processors, cause the processors to perform the data-driven planning and truncation method described above.

[0044] Compared with the prior art, the beneficial effects of the present invention are:

[0045] First, leveraging the advantages of the YOLOv8 model in image recognition through deep learning technology, efficient and rapid mineral layer detection can be achieved. This method not only reduces reliance on manual annotation but also improves annotation quality and consistency through automated processing, thereby ensuring the efficient use of training data. Therefore, the accuracy and detection capability of the trained model are effectively improved, making accurate identification of mineral layers possible during actual mining operations.

[0046] Secondly, this method optimizes the grid image of the fully mechanized mining face, allowing the initial grid size to adaptively adjust based on the actual hammer crusher radius and roadway cross-sectional area. This feature enables more precise ore layer delineation under complex geological conditions, avoiding resource waste and low mining efficiency caused by unreasonable grid sizes. Simultaneously, path simulation based on the obtained raster map data can effectively generate multiple integrated cutting paths. These paths, after optimization and merging, ensure continuity and efficiency during the mining process, reducing overlapping mining and omissions, and improving ore recovery rates. Attached Figure Description

[0047] Figure 1 This is a schematic diagram of the overall method flow of the present invention;

[0048] Figure 2 These are the original images and sample thermal images of the phosphate rock layer dataset of this invention;

[0049] Figure 3 The model of this invention is used to calculate the ore seam cutting boundary of the fully mechanized mining face;

[0050] Figure 4 This is a schematic diagram of the grid map of the present invention;

[0051] Figure 5 This is a schematic diagram of the overall cutting path of the present invention;

[0052] Figure 6 This is a schematic diagram of the overall device structure of the present invention. Detailed Implementation

[0053] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to specific embodiments.

[0054] It should be noted that, unless otherwise defined, the technical or scientific terms used in this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this invention pertains. The terms "first," "second," and similar terms used in this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are used only to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0055] Example:

[0056] Please see Figures 1-5 The present invention provides a technical solution:

[0057] A data-driven planning and truncation method, comprising the following steps:

[0058] Step 1: Obtain the image set data of the phosphate rock layer, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images using a manual annotation scheme. Store the labeled images and the corresponding original images into the database to form the training dataset.

[0059] Acquire phosphate rock layer image set data, and preprocess the acquired images. The preprocessing includes image dehazing, image enhancement and denoising. Wavelet transform is used to denoise each phosphate rock layer image, and bilateral filtering is used to enhance each phosphate rock layer image sample.

[0060] The main purpose of the image dehazing preprocessing is to eliminate image blurring and reduced contrast caused by environmental factors such as haze and dust. These factors can affect the visibility of key features in the image, making the mineral layer identification process difficult. Therefore, dehazing processing can improve image clarity and contrast, making mineral features more obvious and enhancing the accuracy of subsequent image analysis.

[0061] The Retinex algorithm is a dehazing technique based on the human visual perception model. It improves image quality by adjusting contrast and brightness, making details more apparent.

[0062] This method typically involves the multi-scale Retinex algorithm, which uses convolutional kernels of different scales to process images in order to better recover details.

[0063] The method for denoising and enhancing each phosphate rock layer image sample is as follows: The denoising method using wavelet transform is applied to the distortion-corrected image. The specific steps of the wavelet transform denoising method include: decomposing the distortion-corrected image using wavelet transform to obtain wavelet coefficients at different scales and directions; thresholding the wavelet coefficients, setting low-amplitude wavelet coefficients to zero while retaining high-amplitude wavelet coefficients; and performing an inverse transform on the thresholded wavelet coefficients to reconstruct the image, thus completing the image denoising process.

[0064] Bilateral filtering was used to enhance the details of the acquired shrimp images. The specific formula used for the filtering transformation is as follows:

[0065]

[0066] In the formula, The coordinate vector within the image coordinate system. Coordinate vector grayscale value at that location grayscale value The grayscale value after bilateral filtering transformation Both are Gaussian functions, among which The formula used is:

[0067]

[0068]

[0069] In the formula, The coordinate vector within the image coordinate system. Coordinate vector grayscale value at that location and They are respectively The standard deviation.

[0070] The rock strata, phosphate rock strata, and the truncated boundaries between them were labeled on the preprocessed images using a manual annotation scheme. The specific annotation logic was as follows: the open-source software Labelme was used to label the samples, the dataset was converted into JSON format, and it included two categories: phosphate rock strata and rock strata. Each image had a corresponding JSON annotation file, providing object outlines and corresponding classification labels. The dataset contained a total of 2584 images, which were divided into training and validation sets in an 8:2 ratio.

[0071] The method for generating the training dataset is as follows: the labeled images are stored in the database and mapped one by one with the images labeled with the corresponding labels to form a corresponding grid, and the resulting grid is recorded as the training sample dataset.

[0072] Step 2: Based on the data in the training dataset, establish a YOLOv8 detection network model. Use the unlabeled original images of phosphate rock mining in the training dataset as input to the YOLOv8 detection network model, and use the labeled corresponding images in the training dataset as labels to train the model.

[0073] Based on the data in the training dataset, a YOLOv8 detection network model is established. The training method of the model includes a freeze phase and a thaw phase. Finally, the difference between the loss of the training set and the validation set is within 0.016. The training ends. The input of the trained YOLOv8 detection network model is the target mine image, and the output is an image with the rock strata, phosphate rock layer and the cut boundary between them marked in the image.

[0074] The freezing phase refers to selectively fixing the weights of some network layers during the early stages of model training, without updating them. The main purpose of this phase is to utilize the existing knowledge in the pre-trained model, reduce training time, and prevent overfitting during the initial learning process. Typically, the first few layers of the network are frozen. These layers are responsible for extracting basic image features, which are usually similar across different datasets. Only the last few layers are trained. The weights of these layers can be updated according to the current dataset to adapt to the needs of specific tasks. By freezing layers, the computational burden can be reduced, training speed can be improved, and learning instability caused by random initialization can be avoided. Furthermore, the model can converge faster, improving training efficiency.

[0075] The unfreezing phase refers to the gradual unfreezing of previously frozen layers after the freezing phase, allowing them to be updated during training. The main purpose of this phase is to further refine the model's learning, enabling it to fit the current dataset more accurately. In this phase, layer-by-layer unfreezing can be chosen, gradually unlocking the frozen layers to allow end-to-end training of the entire network. Typically, unfreezing can begin from the last layer and proceed towards the preceding layers. Unfreezing allows the model to better adapt to the characteristics of the specific dataset, especially learning details and patterns unique to the dataset. This process helps improve the final performance and accuracy of the model.

[0076] The primary function of the YOLOv8s network is real-time object detection. Object detection is a crucial task in computer vision, aiming to detect various categories of objects from images or videos and accurately label their locations. YOLOv8s, a model in the YOLO (You Only Look Once) family, strikes a balance between real-time performance and accuracy, making it suitable for scenarios requiring high speed and lightweight models.

[0077] Step 3: Real-time acquisition of target mine images and input into the trained YOLOv8 detection network model. Output labeled mining images of the ore layers to be segmented. Based on the labeled contours of the output images, extract the fully mechanized mining face image and divide its region using an initial grid to obtain the fully mechanized mining face grid image.

[0078] This process involves extracting information from each labeled region in the output of the YOLOv8 model, including the bounding box, category, and confidence level of the ore layer. The labeled ore layer regions are analyzed to extract key features, which will provide the foundation for subsequent geological model construction. The labeled ore layer regions are then converted into voxel data, i.e., discrete voxels (small cubes) are created in 3D space to represent the 3D structure of the ore layer. Each voxel can contain characteristic information of the ore layer, such as ore type, content, and physical properties. Geological modeling software or algorithms are then used to integrate the extracted ore layer feature data into a complete geological model. Commonly used software includes Geostudio and Leapfrog Geo.

[0079] Step 4: Based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, optimize the initial grid size, determine the adaptive grid size, and divide the fully mechanized mining face image into regions according to the obtained adaptive grid size to obtain raster map data.

[0080] Based on the acquired hammer crusher radius and area data, the initial mesh size is optimized to determine an adaptive mesh size. The formula used to calculate the adaptive mesh size is as follows:

[0081]

[0082] In the formula, To adapt to the grid size, This is the initial grid size. The factor representing the influence of the damage radius is... This is an index representing the impact of geological complexity, among which... This is the weighting coefficient for the influence of the damage radius. The weighting coefficients for the geological complexity influence index are as follows: , and and Both are greater than 0. The adaptive grid size and initial grid size refer to the grid area determined by the aspect ratio.

[0083] It should be noted that the damage radius influence coefficient This is used to express the influence of the hammer crusher radius on the mesh size, where the crusher radius influence coefficient is... Positive values ​​increase the grid size, while negative values ​​decrease the grid size.

[0084] Geological complexity impact index Used to express the influence of geological features on grid size, where the geological complexity influence index is... Positive values ​​increase the grid size, while negative values ​​decrease the grid size.

[0085] The damage radius directly impacts mining safety: the hammer crusher radius is directly related to the stability of the ore layer and the safety risks during mining. In actual mining, areas with greater damage may lead to more significant geological changes and potential safety hazards. Therefore, assigning a higher weight to the impact of the damage radius helps to handle these high-risk areas more cautiously during grid division. Although geological complexity also affects the mining process, its complexity may exhibit a slower trend of change compared to the direct impact of the damage radius. Therefore, giving lower weight to geological complexity is logical, allowing for reasonable grid division while ensuring safety. , and and All are greater than 0.

[0086] Among them, the influence coefficient of the damage radius The formula used for the calculation is:

[0087]

[0088] In the formula, For reference, the radius of the hammer crusher This represents the actual hammer crusher radius.

[0089] It's important to note that a larger hammer crusher radius means a wider area may be affected during mining. In such cases, to improve safety, a more detailed mesh may be needed for localized areas to accurately monitor and control potential risks. Smaller meshes allow engineers to better monitor and analyze changes in the damaged area, adjust mining strategies promptly, and minimize the impact on the surrounding environment. A larger hammer crusher radius typically means more complex geological structures may be involved during mining. To more accurately reflect these complex geological changes, smaller meshes are usually needed to capture subtle variations and features. Smaller meshes better capture minute differences in the rock mass, such as variations in different rock layers and fractures, ensuring the model reflects the actual geological conditions. Therefore, the impact coefficient of the crusher radius is crucial. Inversely proportional to the actual hammer crusher radius, through It indicates an inverse relationship.

[0090] Among them, the geological complexity influence index The geological complexity influence index is calculated based on the cross-sectional area of ​​the roadway in the fully mechanized mining face and the hardness and number of cracks in the hard rock strata within the roadway cross-section. The formula used for the calculation is:

[0091]

[0092] In the formula, This refers to the cross-sectional area of ​​the roadway in the fully mechanized mining face. For reference, the cross-sectional area of ​​the tunnel, This represents the average hardness of the hard rock strata within the cross-section of the fully mechanized mining face roadway. This represents the number of cracks within the cross-section of the tunnel.

[0093] Among them, the larger the cross-sectional area of ​​the roadway in the fully mechanized mining face, the larger the working area, which allows for an increase in grid size to save computational resources. Therefore, the geological complexity index is affected. Cross-sectional area of ​​the roadway with the fully mechanized mining face Proportional, through It indicates a direct proportional relationship.

[0094] Hard rock formations typically possess more complex geological structures, such as fractures, interlayers, and heterogeneity. These subtle differences can significantly impact the stability of the rock formation and the difficulty of mining, thus requiring smaller meshes to capture these details and ensure the model accurately reflects the true condition of the rock formation. Mining hard rock formations can lead to higher risks, such as rock collapse or fracturing. To better address these risks, reducing the mesh size allows for more refined analysis and prediction of potential safety issues, enabling the development of more effective safety measures. Therefore, geological complexity significantly impacts the index. The average hardness of the hard rock strata within the cross-section of the fully mechanized mining face roadway Inversely proportional, through Logarithmic functions can effectively express the nonlinear relationship that may exist between hardness and mesh size. As hardness increases, its effect on mesh size may gradually decrease; using a logarithmic function can more accurately reflect this complex correlation.

[0095] Cracks, as a significant feature of rock masses, can substantially affect their mechanical properties and stability. Reducing the mesh size allows for better capture of crack geometry, distribution, and quantity. Smaller meshes better reflect the non-uniformity of the rock mass under the influence of cracks, ensuring that the impact of cracks on stress and deformation fields is considered in numerical simulations. Therefore, the geological complexity index is affected. The number of cracks in the cross-section of the tunnel Inversely proportional, through The properties of the square root function gradually reduce the impact of increasing crack number on mesh size, thus avoiding excessive mesh refinement when the crack number is extremely high.

[0096] The fully mechanized mining face image is divided into regions based on the obtained adaptive grid size to obtain raster map data. The specific steps for obtaining raster map data are as follows: the fully mechanized mining face image is divided into regions using the determined adaptive grid size to form several grids. The annotation boxes of phosphate rock, hard rock layer and the cutting boundary between them in each grid are analyzed. Phosphate rock is marked in white and hard rock layer is marked in black to indicate the presence of obstacles. Grids with a black area ratio of more than 40% are marked as black, and grids with a black area ratio of less than 40% are marked as white, thereby generating a raster map.

[0097] Step 5: Based on the obtained raster map data, simulate the cutting path, traverse each raster, generate several path segments, and merge the generated path segments to obtain several comprehensive cutting paths, thus completing the planned cutting.

[0098] After obtaining the raster map, a Cartesian coordinate system is established, with its origin O located at the top left corner of the map, the x-axis pointing downwards, and the y-axis pointing to the right. A matrix of the same dimension as the raster map is created, and the raster to be traversed is represented by the number 0, the traversed raster by the number 1, the raster marked as black by the number 2, and the raster not marked as black by the number 3. Let all the raster cells on the map form a set. Then the grid cells to be traversed can be represented as:

[0099]

[0100] in, This is the row index of the i-th cell in the established Cartesian coordinate system. This is the column index of the i-th grid center point in the established Cartesian coordinate system. This indicates the traversal state of the i-th grid cell. This indicates that the grid cell is to be traversed, where For the index of the raster, , This represents the total number of grid cells.

[0101] Based on the obtained raster map data, the truncation path is simulated. Each raster is traversed to generate several path segments, defined as follows: a path segment is an ordered set of all connected raster cells in the same row on the map that are not blocked by obstacles. A path segment in column R then satisfies... Specifically, it satisfies:

[0102]

[0103] In the formula, This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are no obstacles in the grid at the row position. The difference between consecutive raster indices within a path segment is 1, used to ensure they are consecutive. For the index of the vertical raster in column R, This represents the total number of vertical grid cells in column R. For the index in column R, excluding the first vertical raster;

[0104] If a certain region of a map is divided into n path segments, then the set of these segments is... And satisfy:

[0105]

[0106] In the formula, Let f(x) represent any two sets of path segments, provided that they have no intersection and are independent of each other.

[0107] The specific logic for generating and merging several path segments is as follows: A start and end point are randomly generated within a certain area, and a path is generated. Based on specific rules such as optimal path algorithms and terrain features, path segments are generated. The steps for merging path segments are as follows: Check for overlap: First, check whether the path segments intersect or overlap. Spatial data structures can be used to accelerate the search process. Detect adjacency: Check whether the endpoints of the path segments are adjacent, i.e., whether the end point of one path segment is very close to the start point of another path segment.

[0108] Based on the detection results of intersection and adjacency, select a merging strategy:

[0109] Endpoint merging: If the endpoints of two path segments are adjacent, they can be directly merged into one path.

[0110] Overlapping merging: For overlapping path segments, the overlapping area can be found to form a new path segment.

[0111] Smoothing: After merging, a smoothing algorithm can be used to optimize the shape of the new path, making it smoother. The merged path segments are then organized into a composite truncated path: Path sorting: The composite path is sorted according to certain rules such as direction and distance. Output format: The composite truncated path is converted into the required data format as needed for subsequent analysis and visualization.

[0112] Choosing a specific path involves calculating the total time required for the excavating loading robot to complete the path, and selecting the path with the shortest time to execute the task. The specific formula used to calculate the total time required for the movement is as follows:

[0113]

[0114] In the formula, Indicates starting from the origin Move to the finish line Total time required The equivalent path length. The average speed of the excavating loading robot as it moves in a straight line. This represents the total number of turns during the movement. Each turn requires time.

[0115] Please participate Figure 6 The present invention also provides a data-driven planning and truncation device, which is used to execute the above-described data-driven planning and truncation method, comprising:

[0116] The training sample processing module is used to acquire phosphate rock layer image set data, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images. The labeled images and the corresponding original images are stored in the database to form the training dataset.

[0117] The prediction model segmentation module is used to build a YOLOv8 detection network model based on the data in the training dataset. The original images of the unlabeled phosphate rock mining layer in the training dataset are used as input to the YOLOv8 detection network model, and the corresponding labeled images in the training dataset are used as labels to train the model.

[0118] The planar image processing module is used to acquire images of the target mine in real time and input them into the trained YOLOv8 detection network model. It outputs the mining images of the ore layer to be segmented with labels. Based on the label contours of the output images, it extracts the fully mechanized mining face image and divides its region through an initial grid to obtain the fully mechanized mining face grid image.

[0119] The grid size optimization module is used to optimize the initial grid size based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, determine the adaptive grid size, and divide the fully mechanized mining face image into regions according to the obtained adaptive grid size to obtain raster map data.

[0120] The cut path synthesis module is used to simulate the cut path based on the obtained raster map data, traverse each raster, generate several path segments, and merge the generated path segments to obtain several comprehensive cut paths, thus completing the planned cut.

[0121] The present invention also provides a non-volatile computer-readable storage medium containing computer-executable instructions that, when executed by one or more processors, cause the processors to perform the data-driven planning and truncation method described above.

[0122] The above formulas are all dimensionless calculations. The formulas are derived from software simulations based on a large amount of collected data to obtain the most recent real-world results. The preset parameters in the formulas are set by those skilled in the art according to the actual situation.

[0123] The above embodiments can be implemented, in whole or in part, by software, hardware, firmware, or any other combination thereof. When implemented in software, the above embodiments can be implemented, in whole or in part, as a computer program product. Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented by electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution.

[0124] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0125] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any changes or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application.

Claims

1. A data-driven planning and cutting method, characterized in that, The specific steps include: Acquire phosphate rock layer image set data, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images. Store the labeled images and the corresponding original images in the database to form a training dataset. Based on the data in the training dataset, a YOLOv8 detection network model is established. The original images of phosphate rock mining without labels in the training dataset are used as input to the YOLOv8 detection network model, and the corresponding labeled images in the training dataset are used as labels to train the model. Real-time acquisition of target mine images is input into the trained YOLOv8 detection network model, outputting a labeled mining image of the ore layer to be segmented. Based on the labeled contour of the output image, the fully mechanized mining face image is extracted, and its region is divided by an initial grid to obtain a fully mechanized mining face grid image. Based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, the initial grid size is optimized to determine the adaptive grid size. The fully mechanized mining face image is then divided into regions according to the obtained adaptive grid size to obtain raster map data. Based on the obtained raster map data, the cutting path is simulated, each raster is traversed, several path segments are generated, and the generated path segments are merged to obtain several comprehensive cutting paths, thus completing the planned cutting. Based on the acquired hammer crusher radius and area data, the initial mesh size is optimized to determine an adaptive mesh size. The formula used to calculate the adaptive mesh size is as follows: In the formula, To adapt to the grid size, This is the initial grid size. The factor representing the influence of the damage radius is... This is an index representing the impact of geological complexity, among which... This is the weighting coefficient for the influence of the damage radius. The weighting coefficients for the geological complexity influence index are as follows: , and and All are greater than 0; Among them, the influence coefficient of the damage radius The formula used for the calculation is: In the formula, For reference, the radius of the hammer crusher This is the actual hammer crusher radius; Among them, the geological complexity influence index The geological complexity influence index is calculated based on the cross-sectional area of ​​the roadway in the fully mechanized mining face and the hardness and number of cracks in the hard rock strata within the roadway cross-section. The formula used for the calculation is: In the formula, This refers to the cross-sectional area of ​​the roadway in the fully mechanized mining face. For reference, the cross-sectional area of ​​the tunnel, This represents the average hardness of the hard rock strata within the cross-section of the fully mechanized mining face roadway. This represents the number of cracks within the cross-section of the tunnel.

2. The data-driven planning and truncation method according to claim 1, characterized in that: Acquire phosphate rock layer image set data, and preprocess the acquired images. The preprocessing includes image dehazing, image enhancement and denoising. Wavelet transform is used to denoise each phosphate rock layer image, and bilateral filtering is used to enhance each phosphate rock layer image sample.

3. The data-driven planning and truncation method according to claim 2, characterized in that: The rock strata, phosphate rock strata, and the truncated boundaries between them were labeled on the preprocessed images using a manual annotation scheme. The specific annotation logic was as follows: the open-source software Labelme was used to label the samples, the dataset was converted into JSON format, and it contained two categories: phosphate rock strata and rock strata. Each image had a corresponding JSON annotation file, providing object outlines and corresponding classification labels. The dataset contained a total of 2584 images, which were divided into training set and validation set in an 8:2 ratio. The method for generating the training dataset is as follows: the labeled images are stored in the database and mapped one by one with the images labeled with the corresponding labels to form a corresponding grid, and the resulting grid is recorded as the training sample dataset.

4. The data-driven planning and truncation method according to claim 3, characterized in that: Based on the data in the training dataset, a YOLOv8 detection network model is established. The training method of the model includes a freeze phase and a thaw phase. Finally, the difference between the loss of the training set and the validation set is within 0.

016. The training ends. The input of the trained YOLOv8 detection network model is the target mine image, and the output is an image with the rock strata, phosphate rock layer and the cut boundary between them marked in the image.

5. The data-driven planning and truncation method according to claim 4, characterized in that: The fully mechanized mining face image is divided into regions based on the obtained adaptive grid size to obtain raster map data. The specific steps for obtaining raster map data are as follows: the fully mechanized mining face image is divided into regions using a determined adaptive grid size to form several grids. The annotation boxes of phosphate rock, hard rock layer and the cutting boundary between them in each grid are analyzed. Phosphate rock is marked in white and hard rock layer is marked in black to indicate the presence of obstacles. Grids with a black area ratio of more than 40% are marked as black, and grids with a black area ratio of less than 40% are marked as white, thereby generating a raster map.

6. The data-driven planning and truncation method according to claim 5, characterized in that: After obtaining the raster map, a Cartesian coordinate system is established, with its origin O located at the top left corner of the map, the x-axis pointing downwards, and the y-axis pointing to the right. A matrix of the same dimension as the raster map is created, and the raster to be traversed is represented by the number 0, the traversed raster by the number 1, the raster marked as black by the number 2, and the raster not marked as black by the number 3. Let all the raster cells on the map form a set. Then the grid cells to be traversed can be represented as: in, This is the row index of the i-th cell in the established Cartesian coordinate system. This is the column index of the i-th grid center point in the established Cartesian coordinate system. This indicates the traversal state of the i-th grid cell. This indicates that the grid cell is to be traversed, where For the index of the raster, , This represents the total number of grid cells. Based on the obtained raster map data, the truncation path is simulated. Each raster is traversed to generate several path segments, defined as follows: a path segment is an ordered set of all connected raster cells in the same row on the map that are not blocked by obstacles. A path segment in column R then satisfies... Specifically, it satisfies: In the formula, This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are obstacles at the row grid. This indicates that in column R, the first... There are no obstacles in the grid at the row position. The difference between consecutive raster indices within a path segment is 1, used to ensure they are consecutive. For the index of the vertical raster in column R, This represents the total number of vertical grid cells in column R. Let R be the index of the first vertical grid cell in column R; where a map's traversable region is divided into n path segments, then the set of these segments is... And satisfy: In the formula, Let f(x) represent any two sets of path segments, provided that they have no intersection and are independent of each other.

7. A data-driven planning and cutting device, characterized in that: The data-driven planning and cutting device is used to execute the data-driven planning and cutting method according to any one of claims 1-6, comprising: The training sample processing module is used to acquire phosphate rock layer image set data, preprocess the acquired images, and label the rock layer, phosphate rock layer and the truncated boundary between them on the preprocessed images. The labeled images and the corresponding original images are stored in the database to form the training dataset. The prediction model segmentation module is used to build a YOLOv8 detection network model based on the data in the training dataset. The original images of the unlabeled phosphate rock mining layer in the training dataset are used as input to the YOLOv8 detection network model, and the corresponding labeled images in the training dataset are used as labels to train the model. The planar image processing module is used to acquire images of the target mine in real time and input them into the trained YOLOv8 detection network model. It outputs the mining images of the ore layer to be segmented with labels. Based on the label contours of the output images, it extracts the fully mechanized mining face image and divides its region through an initial grid to obtain the fully mechanized mining face grid image. The grid size optimization module is used to optimize the initial grid size based on the breaking radius of the hydraulic breaker and the area of ​​the roadway cross section of the fully mechanized mining face, determine the adaptive grid size, and divide the fully mechanized mining face image into regions according to the obtained adaptive grid size to obtain raster map data. The cut path synthesis module is used to simulate the cut path based on the obtained raster map data, traverse each raster, generate several path segments, and merge the generated path segments to obtain several comprehensive cut paths, thus completing the planned cut.

8. A non-volatile computer-readable storage medium containing computer-executable instructions, characterized in that: When the computer-executable instructions are executed by one or more processors, the processors perform a data-driven planning and truncation method according to any one of claims 1 to 6.

Images